A method, recording medium and system for regulating river ecological flow

Abstract

The present invention belongs to the technical field of hydrology and water resources management, and discloses a method for regulating the ecological flow of a river channel. The method collects time series data of the exposed area of ​​sandbars in the river channel and the runoff of fixed measuring point sections; uses a jump detection algorithm to determine two intermediate values ​​and divide the intensity of wind and sand activity into three sections; establishes function models respectively, and selects to use a particle swarm algorithm or a Bayesian algorithm to optimize the model parameters according to the adaptability of different models; uses an adaptive dynamic sliding window algorithm to continuously sample the time series data, inputs the data into the model calculation, and obtains a reference value for the minimum ecological flow. The method integrates technologies in the fields of hydrology, meteorology, geomorphology, and artificial intelligence to establish a linkage model of "exposed area of ​​sandbars-runoff-wind and sand probability", solves the problem of cross-temporal and spatial decoupling, and is suitable for the collaborative optimization control process of ecological flow and wind and sand suppression. The present invention also provides a non-transient readable recording medium storing the program of the method and a system containing the medium, which can call the program through a processing circuit to execute the above method.

Description

Technical Field

[0001] The invention belongs to the technical field of hydrology and water resources management, and discloses a method, recording medium and system for regulating river ecological flow. Background Art

[0002] Dust sources, strong winds, and unstable atmospheric circulation are three key factors in the generation of sandstorms. Wind conditions are particularly crucial for sandstorm activity, with significant variations in wind speed, direction, and the probability of dust-bearing winds across different regions. Understanding the characteristics and changing patterns of these wind conditions is crucial for accurately assessing sandstorm probability and developing targeted sandstorm control measures. However, current monitoring and prediction models for sandstorm probability are often limited to specific regions and seasons and lack universal applicability. Even within the same region and time period, the impact of varying windstorm intensities is difficult to fit into a function curve, resulting in change point detection issues and failing to meet the growing demand for sandstorm control.

[0003] Water ecological flow is essential for maintaining the structure and function of aquatic ecosystems such as rivers and lakes. It not only ensures the stability of river channels, prevents dry-up, enhances river self-purification, and replenishes groundwater, but also maintains the water volume and water levels needed by aquatic organisms during the dry season, thus playing a crucial role in maintaining aquatic ecological balance.

[0004] However, due to limitations in natural endowments, irrational development and utilization, and global climate change, conflicts between domestic, productive, and ecological water use have become prominent in some river basins, making it difficult to ensure the ecological flow of rivers and lakes. This has led to a series of serious problems, including river dry-ups, lake shrinkage, biodiversity loss, and a decline in ecological services. For example, the decline of water resources in the Yellow River basin and the increase in economic and social use have led to an imbalance in water supply and demand, reduced river runoff, weakened hydrodynamics, and difficulties in ensuring the basic sediment transport water required to maintain the functions of sediment-laden rivers. How to scientifically and rationally determine ecological flow targets and effectively guarantee them has become a key issue that needs to be addressed in the current fields of water resources management and water ecological protection.

[0005] Currently, relatively little research exists on the relationship between windstorm probability and water ecological flow, with the two often studied as independent fields. However, in reality, they are closely linked. Windstorm activity alters the properties of the underlying surface, affecting surface runoff and soil evaporation, which in turn influences the distribution and circulation of water resources, indirectly impacting water ecological flow. Conversely, changes in water ecological flow can also affect vegetation growth. Vegetation, a key factor influencing windstorm activity, can in turn affect windstorm probability. For example, in arid and semi-arid regions, reduced river flow leads to the degradation of coastal vegetation, exacerbating desertification and exposing bare sandbars within river channels, increasing the probability of windstorms.

[0006] Sudden events can also affect the probability of sandstorms and water ecological flows. For example, building dams or diversion projects on rivers, or planting trees to prevent wind and sand, means the interaction between sandstorm probability and water ecological flows is not a simple functional relationship. The factors involved are complex and intricate. Directly fitting the relationship between sandstorm probability and water ecological flows using artificial intelligence (AI) will miss key features, resulting in underfitting and a failure to accurately reflect the relationship. Complex models will overfit and lack generalization. This is similar to determining whether chemotherapy has a positive or negative effect on cancer treatment; AI modeling cannot accurately fit the relationship between the two. The lack of effective comprehensive analytical methods and technical means limits our comprehensive understanding of the overall state of the regional ecological environment and its effective protection. Summary of the Invention

[0007] The inventive concept of this application is to find intermediate parameters that are directly related to water ecological flow and sandstorm probability, establish a segmented model, reveal the indirect relationship between water ecological flow and sandstorm probability, and ultimately improve the impact of sandstorm activities on humans based on the allocation of water resources. This requires overcoming many obstacles in the cross-disciplinary docking process. The present invention provides a method for regulating river ecological flow, including the following steps:

[0008] S1. Collect time series data on the exposed area of ​​sandbars in the river channel and the probability of wind-blown sand activity at fixed measuring point sections;

[0009] S2. Establish a function model between the exposed area of ​​the sandbar and the runoff volume of the fixed measuring point section at the same time, and use the particle swarm algorithm to optimize the parameters of the function model;

[0010] S3. Establish a sandstorm activity intensity model, determine the intermediate value between two sandstorm activity intensities using a jump detection algorithm, and classify the sandstorm activity intensity into three intervals: mild, moderate, and severe based on the two intermediate values. The jump detection algorithm is a combination of a pruned linear time exact algorithm and a change point detection algorithm, and is used to identify the nonlinear structure of the sandstorm activity intensity time series.

[0011] S4. Establish a sandstorm response model for each of the three intervals, describing the relationship between the probability of sandstorm activity and the exposed area of ​​sandbars within the same time period. For the continuous sampling of time series data involved in the function model, an adaptive dynamic sliding window algorithm is used. The window length is dynamically adjusted according to the intensity of sandstorm activity, ranging from 2 to 4 months, with an initial sliding step of 10 days.

[0012] S5. Determine the critical sandbar exposed area according to the set threshold of sand and wind activity probability in the sand and wind response model, substitute the critical exposed area into the function model in step S2, and calculate the corresponding runoff of the fixed measurement point section as the minimum ecological flow; adjust the sliding step length according to the change range of the minimum ecological flow within the set time period;

[0013] S6. Manually adjust the runoff of the fixed measurement point section of the river channel above the minimum ecological flow and update the model data for optimization in real time.

[0014] Preferably, the sand and wind activity intensity model is: set the sand and wind activity intensity SWAD = α·FD + β·BD + γ·DS + δ·TSD; where, FD is the floating dust frequency, BD is the sand blowing frequency, DS is the sandstorm frequency, TSD is the dust storm duration, and α, β, γ, δ are seasonal weight coefficients, determined by the principal component analysis method in the same season; set the intermediate value separating the mild and moderate intervals as SWAD_L, and the intermediate value separating the moderate and severe intervals as SWAD_H; then the three sand and wind response models:

[0015] In the mild interval, the sand and wind response model corresponding to the state of SWAD ≤ SWAD_L:

[0016] P = k1·A + d1;

[0017] In the moderate interval, the sand and wind response model corresponding to the state of SWAD_L < SWAD ≤ SWAD_H: [[ID=十六]] [[ID=十七]]

[0018] P = k2·A 2 + k3·A + d2;

[0019] In the severe interval, the sand and wind response model corresponding to the state of SWAD > SWAD_H:

[0020] [[ID=2十六]]P = k4·e λA + d3;

[0021] Among them, λ is the exponential growth rate coefficient, representing the rate of sand and wind outbreaks, k1, k2, k3, k4 are proportionality coefficients, d1, d2, d3 are constant terms, optimized by the Bayesian algorithm, P is the sand and wind activity probability, and A is the sandbar exposed area, which belongs to the observed data.

[0022] Preferably, the function model between the sandbar exposed area in step S2 and the runoff of the fixed measurement point section at the same moment is:

[0023] When the sandbar exposed area A ≤ ζ·A sand it is determined that the sandbar is completely submerged, and it is default that the ecological flow does not need to be adjusted;

[0024] Set the seasonal exposed sandbar area threshold When ζ·Asand ≤A≤A thres When the cross-sectional runoff at the fixed measuring point is Q=Q base +κ·(A-ζ·A sand )

[0025] If A>A thres , fixed measuring point cross-sectional runoff

[0026] in:

[0027] A sand : the total area of ​​exposed sandbars, determined by the maximum historical exposed area extracted from remote sensing images;

[0028] ζ: flooding threshold coefficient, default 0.02, optimization range 0.01-0.05;

[0029] Q base : Basic ecological flow, taking the 90% quantile of the historical minimum monthly average flow;

[0030] κ: flow compensation coefficient, determined by linear regression;

[0031] a is the maximum exposed area coefficient, which represents the theoretical maximum exposed area under zero flow conditions;

[0032] b is the attenuation coefficient, which represents the rate of reduction of the exposed area when the flow rate increases;

[0033] c is the shape parameter, which determines the nonlinearity of the function curve;

[0034] μ is the average baseline exposed area;

[0035] ω is the seasonal variation coefficient, which represents the fluctuation range of exposed area relative to the reference area;

[0036] φ is the phase parameter, which characterizes the time offset of seasonal variation;

[0037] t is the time point of observation, and the serial number sorted from 1 to 365 within the year is used as the t value;

[0038] Parameters a, b, c, μ, ω, and φ are optimized by particle swarm optimization, and their hyperparameters are:

[0039] Goodness of fit R 2 ≥0.90; the root mean square error of the sandbar exposed area A is less than 10% of the average sandbar exposed area.

[0040] Preferably, the step of collecting time series data of exposed areas of sandbars in a river channel comprises:

[0041] The spatial resolution of the selected remote sensing camera should be ≤30×30m 2, covering at least three hydrological years with wet, normal, and dry periods. When the cloud cover rate of the images was ≤5%, remote sensing images of the river area were collected; radiometric calibration and atmospheric correction were performed on the remote sensing images using ENVI software, and the normalized difference vegetation index (NDVI) was combined with supervised classification using a deep residual convolutional neural network to extract the vector boundaries of exposed sandbars in each period and calculate the exposed sandbar area A; NDVI = (NIR-R) / (NIR+R), where NIR is the reflectance of the near-infrared band and R is the reflectance of the red light band; the standard for identifying exposed sandbars was NDVI < 0.2 and the terrain elevation was between the lowest water level and the flood level of the river; field measurements were conducted to verify whether the accuracy of the exposed sandbar area extraction met the requirements.

[0042] Preferably, the deep residual convolutional neural network is based on the ResNet-50 architecture, which contains 5 residual blocks and a total of 50 convolutional layers; a channel attention module is added to the residual block to improve the sensitivity to sandbar features; a feature pyramid structure is adopted, combined with feature maps of different scales, to improve the detection ability of sandbars of different sizes; a boundary refinement network is added to accurately extract sandbar boundaries; a combination of weighted cross entropy loss and boundary perception loss is used to improve the accuracy of sandbar boundary recognition; among them, the neural network training parameters include learning rate: initial value 0.001, adjusted by cosine annealing strategy; an adaptive moment estimation optimizer is used, the exponential decay rate of the first-order moment β1=0.9, and the exponential decay rate of the second-order moment β2=0.999; the batch size is 16, and the number of training rounds is initially set to 200 rounds. If the verification accuracy does not improve after 10 rounds, the training is completed in advance.

[0043] Preferably, when the cloud coverage is >5% but <20%, a cloud detection and cloud removal algorithm (Fmask algorithm) is used, and the area under the cloud is supplemented by time series interpolation; when there is data missing in a local area, a spatiotemporal fusion algorithm is used to fill the data by combining the data of adjacent time points; when data is completely missing for a specific period, a Markov chain Monte Carlo method based on historical data of the same period is used to generate simulated data.

[0044] Preferably, the window length in months is: WL=2+2×(SWAD-SWAD_min) / (SWAD_max-SWAD_min); wherein, SWAD_min is the minimum SWAD value of historical observations; SWAD_max is the maximum SWAD value of historical observations; when the minimum ecological flow calculated twice in a row changes by more than 20%, the step length is reduced to 5 days; when the minimum ecological flow calculated for 4 consecutive times changes by less than 5%, the step length is increased to 15 days; when the minimum ecological flow Q_eco calculated twice in a row has opposite change trends and the amplitude exceeds 15%, the change point detection is triggered to determine whether it is a real change in hydrological and meteorological conditions or an abnormal fluctuation. When it is determined to be a real change, the manual verification process is started; when it is determined to be an abnormal fluctuation, the smoothing processing technology is used to reduce the impact of the fluctuation.

[0045] Preferably, a dynamic model update mechanism is also included: the improvement of wind and sand activity after the implementation of ecological flow regulation is evaluated every quarter; when the SWAD value of wind and sand activity intensity is monitored to drop by more than 15% for two consecutive quarters, the SWAD_L and SWAD_H values ​​and wind and sand response model parameters are automatically adjusted dynamically using the latest monitoring data; through the validation set test, when the prediction accuracy is improved by more than 5%, the model parameters are updated, otherwise the original model is retained, the new data is included in the training set, and wait for the next update; when the cumulative amount of new observation data exceeds 25% of the original training set or the environmental conditions change significantly, all models are fully updated.

[0046] Another embodiment of the present invention is to provide a non-transitory readable recording medium for storing one or more programs containing a plurality of instructions. When the instructions are executed, the processing circuit will execute the above-mentioned method for regulating river ecological flow.

[0047] Another embodiment of the present invention is to provide a system for regulating the ecological flow of a river, comprising a processing circuit and a memory electrically coupled thereto, wherein the memory is configured to store at least one program, wherein the program includes a plurality of instructions, and the processing circuit runs the program to execute the above-mentioned method for regulating the ecological flow of a river.

[0048] Compared with the prior art, the method, recording medium and system for regulating river ecological flow provided by the present invention have the following beneficial effects:

[0049] The proposed Seasonal Sand Activity Characteristic Value (SWAD) evaluation system breaks through the limitations of a single indicator and enables differentiated quantification and comparison of sand activity intensity. By integrating multiple indicators such as floating dust, blowing sand, and sandstorms, a unified standard is established, enabling horizontal comparison of sand risk across different regions and time periods.

[0050] Multi-threshold response model: This approach proposes a segmented modeling approach for sandstorm activity intensity, addressing the inability of traditional single models to accurately capture the nonlinear relationship between varying sandstorm intensity and exposed sandbar area. This approach automatically adapts mathematical models to the characteristics of varying sandstorm intensity ranges, significantly improving prediction accuracy in critical sandstorm outbreak zones.

[0051] A jump detection algorithm, based on the combination of the Pruned Linear Time Exact (PELT) algorithm and the change point detection algorithm, is used to identify the nonlinear structure of time series of wind and sand activity intensity values. This algorithm automatically identifies wind and sand activity thresholds, eliminating the limitations of traditional methods that rely on expert experience and subjective judgment. This technology can accurately capture the transition points of wind and sand activity patterns, providing a scientific basis for differentiated management.

[0052] A dynamic time window algorithm driven by wind and sand intensity was proposed, overcoming the limitations of fixed windows that cannot adapt to seasonal variations in wind and sand activity and sudden events. This method automatically adjusts the calculation time window and step size based on the intensity of wind and sand activity, improving the timeliness and accuracy of ecological flow calculations.

[0053] Composite modeling of river channel-wind-sand response: This breaks away from the traditional simple power function relationship between flow and area by introducing a seasonal correction term, enabling dynamic modeling of the exposed sandbar area's response to changes in river flow. This composite function simultaneously considers the influence of hydrological processes and seasonal cyclicity, improving area prediction accuracy.

[0054] To address the difficulty in extracting sandbar area, the deep residual convolutional neural network was improved. Through innovations such as attention mechanism and boundary enhancement, the problems of blurred sandbar boundaries, variable scales and complex background were solved, greatly improving the accuracy of sandbar area calculation.

[0055] By integrating technologies from the fields of hydrology, meteorology, geomorphology and artificial intelligence, a closed-loop linkage system of "exposed area-runoff-wind and sand probability" was established, which solved the problem of cross-temporal and spatial decoupling of various state parameters and achieved coordinated optimization of ecological flow and wind and sand suppression. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a diagram of the framework of the "sandbar exposed area-runoff-wind sand probability" model of the present invention;

[0057] Figure 2 This is the data flow diagram of the model of the present invention;

[0058] Figure 3 This is the nonlinear response model of sandbar exposed area-runoff in the embodiment of the present invention. DETAILED DESCRIPTION

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be described below in conjunction with the accompanying drawings. The described embodiments are part of the embodiments of the present invention, but not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without making any innovative efforts shall fall within the scope of protection of the present invention.

[0060] like Figure 1-3 As shown, an embodiment of a method for regulating river ecological flow provided by the present invention is as follows:

[0061] 1. Interpretation of multi-time series remote sensing data

[0062] (1) Remote sensing data acquisition and preprocessing: The spatial resolution of the high-resolution remote sensing image should be ≤30×30m 2 , the time coverage is at least 3 complete hydrological years (including wet, normal and dry years), and the cloud coverage rate of the image is less than 5%;

[0063] ENVI software was used to perform radiometric calibration and atmospheric correction on remote sensing images. An improved deep residual convolutional neural network (DR-CNN) was used to extract the exposed area of ​​sandbars. The classification accuracy was verified by the confusion matrix (Kappa coefficient ≥ 0.80).

[0064] Deep Residual Convolutional Neural Network (DR-CNN) architecture: Based on the ResNet-50 architecture, it contains 5 residual blocks and a total of 50 convolutional layers;

[0065] Introducing the attention mechanism: Adding a channel attention module (Squeeze-and-Excitation module) to the residual block to improve sensitivity to sandbar features;

[0066] Multi-scale feature fusion: Using a feature pyramid structure (FPN) to combine feature maps of different scales, the detection capability of sandbars of different sizes is improved;

[0067] Boundary Enhancement Module: Adds Boundary Refinement Network (BRN) to accurately extract sandbar boundaries;

[0068] Loss function optimization: A combination of weighted cross entropy loss and boundary-aware loss is used to improve the accuracy of sandbar boundary recognition.

[0069] Network training parameters: the initial learning rate is 0.001, and the cosine annealing strategy is used for adjustment; the adaptive moment estimation (Adam) optimizer is used, with β1 = 0.9, β2 = 0.999; the batch size is 16; the number of training rounds is 200, and the early stopping threshold is "no improvement in verification accuracy after 10 rounds".

[0070] When the cloud coverage is >5% but <20%, a cloud detection and cloud removal algorithm (Fmask algorithm) is used, and the area under the cloud is supplemented by time series interpolation. When there is data missing in a local area, a spatiotemporal fusion algorithm (ESTARFM) is used to fill the gap by combining data from adjacent time points. When data is completely missing for a specific period, a Markov Chain Monte Carlo (MCMC) method based on historical data from the same period is used to generate simulated data.

[0071] (2) Extraction method of exposed area of ​​sandbar:

[0072] The Normalized Difference Vegetation Index (NDVI) was combined with supervised classification to extract the vector boundaries of exposed sandbars at each period and calculate the area value A_t.

[0073] NDVI calculation formula: NDVI = (NIR-R) / (NIR+R), where NIR is the near-infrared band reflectance and R is the red light band reflectance;

[0074] Identification criteria for exposed sandbars: NDVI < 0.2 and terrain elevation between the lowest water level and flood level of the river;

[0075] The accuracy requirement for extracting the exposed area of ​​sandbars is: relative error <5%, verified through field verification points.

[0076] 2. Modeling the relationship between exposed sandbar area and runoff volume

[0077] (1) Synchronously collect the river section flow data Q_t at time point t, with a time resolution of ≤10 days and a flow measurement accuracy of ≤±3%;

[0078] (2) Establish a nonlinear response model of the exposed area A of the sandbar and the flow rate Q of the river section:

[0079] A=f(Q)=a·e^(-b·Q^c)+ω·sin(2π·t / 365+φ);

[0080] Since the calculation of minimum ecological flow requires deducing flow from exposed area, an inverse relationship conversion is required:

[0081] (a) When the exposed area of ​​the sandbar A≤ζ·A sand When the sandbar is completely submerged, it is determined that there is no need to adjust the ecological flow by default;

[0082] (b) Setting a threshold for seasonal exposed sandbar area When ζ·A sand ≤A≤A thres When the cross-sectional runoff at the fixed measuring point is Q=Q base +κ·(A-ζ·Asand );

[0083] (c) If A>A thres , fixed measuring point cross-sectional runoff

[0084] Where: A is the exposed area of ​​sandbar (km 2 ); Q is the flow rate of the river section (m3 / s); A sand : total area of ​​sandbar (determined by the maximum historical exposed area extracted from remote sensing images); Q base :Basic ecological flow (m 3 / s), taking the 90% quantile of the historical minimum monthly average flow; ζ: flooding threshold coefficient (dimensionless, default 0.02, optimization range 0.01-0.05); κ: flow compensation coefficient (m 3 / s / m 2 ), determined by linear regression; a is the maximum exposed area coefficient, which represents the theoretical maximum exposed area under zero flow conditions; b is the attenuation coefficient, which controls the rate of decrease of the exposed area when the flow rate increases; c is the shape parameter, which determines the nonlinearity of the curve; ω is the seasonal amplitude area, which represents the amplitude of the seasonal variation of the exposed area; φ is the phase parameter, which controls the time offset of the seasonal variation; t is the time point of observation, and the serial number sorted from 1 to 365 within the year is used as the t value; the parameters a, b, c, φ, ω, and φ are fitted by the particle swarm optimization (PSO) algorithm.

[0085] Fitting requirements: goodness of fit R 2 ≥0.90; RMSE should be less than 10% of the average exposed area; the width of the 95% confidence interval of each parameter should not exceed 30% of the parameter estimate; Particle swarm optimization algorithm parameter settings: population size: 50; number of iterations: 1000; inertia weight: 0.7; individual learning factor: 1.5; global learning factor: 1.5; termination condition: the improvement of the optimal solution for 50 consecutive iterations is <0.1% or the maximum number of iterations is reached.

[0086] 3. Determination of the intensity threshold of wind and sand activity

[0087] (1) Introducing the seasonal sand activity intensity value (SWAD) characterization system:

[0088] SWAD=α·FD+β·BD+γ·DS+δ·TSD;

[0089] Where: FD is the floating dust frequency (d), which refers to the PM10 concentration > 150 μg / m per unit time. 3 and the duration is ≥2 hours;

[0090] BD is the blowing sand frequency (d), which refers to the number of days per unit time when the horizontal visibility is <5000 m and lasts for ≥2 hours;

[0091] DS is the frequency of sandstorms (d), which refers to the number of days per unit time with horizontal visibility <1000 m and duration ≥30 minutes;

[0092] TSD is the duration of sandstorms (h), which refers to the cumulative duration of all sandstorm events in a unit of time in hours;

[0093] α, β, γ, and δ are weight coefficients, which are determined by principal component analysis (PCA).

[0094] Specific method for determining weights:

[0095] 1) Standardize the data of each indicator: Z = (X-μ) / σ, where X is the original indicator value, μ is the mean value, and σ is the standard deviation;

[0096] 2) Calculate the indicator correlation matrix R;

[0097] 3) Perform eigenvalue decomposition on R to obtain the eigenvalue λ and eigenvector V;

[0098] 4) Calculate the principal component contribution rate based on the load coefficient of the first principal component and obtain the weight coefficient after normalization, that is, α+β+γ+δ=1;

[0099] The initial weight is determined by the regional characteristics and the importance of historical sandstorm events with reference to the range of weight coefficients in different seasons. The range of weight coefficients in different seasons is as follows:

[0100] Winter: α∈[0.10,0.20], β∈[0.20,0.30], γ∈[0.30,0.40], δ∈[0.20,0.30];

[0101] Spring: α∈[0.15,0.25],β∈[0.25,0.35],γ∈[0.25,0.35],δ∈[0.15,0.25];

[0102] Summer: α∈[0.25,0.35],β∈[0.25,0.35],γ∈[0.15,0.25],δ∈[0.15,0.25];

[0103] Autumn: α∈[0.20,0.30],β∈[0.20,0.30],γ∈[0.20,0.30],δ∈[0.20,0.30];

[0104] (2) Constructing a multi-level wind and sand response model:

[0105] The intensity of wind and sand activity is divided into three intervals based on the SWAD value, and different fitting functions are used for each interval:

[0106] ① Mild wind and sand activity range (SWAD≤SWAD_L):

[0107] Use the linear equation: P = k1·A + d1;

[0108] This function is applicable to the low-intensity stage when the wind and sand activity is linearly positively correlated with the exposed area, where:

[0109] P is the probability of wind and sand activity (a value between 0 and 1);

[0110] A is the exposed area of ​​sandbar (km 2 );

[0111] k1 is the linear slope coefficient, which represents the increase in the probability of wind and sand caused by the increase in exposure per unit area;

[0112] d1 is a constant term, representing the probability of basic wind and sand activity;

[0113] Recommended parameter ranges: k1∈[0.001,0.005], d1∈[0.05,0.15];

[0114] The lower limit of the parameter represents the area insensitive to wind and sand, and the upper limit represents the area highly sensitive to wind and sand.

[0115] ② Moderate wind and sand activity range (SWAD_L <SWAD≤SWAD_H):

[0116] Use the quadratic polynomial equation: P = k2·A2+k3·A+d2;

[0117] This function can capture the nonlinear characteristics of the probability of wind and sand activity accelerating with the increase of exposed area, where:

[0118] k2 is the quadratic term coefficient, which characterizes the intensity of the nonlinear acceleration effect;

[0119] k3 is the coefficient of the first-order term;

[0120] d2 is a constant term;

[0121] Recommended parameter ranges: k2∈[0.00005,0.0002], k3∈[-0.01,0.01], d2∈[0.1,0.3];

[0122] k2 is always positive to ensure the acceleration effect; k3 can be positive or negative and is determined according to the characteristics of the specific river section.

[0123] ③ Severe wind and sand activity range (SWAD>SWAD_H):

[0124] Using the exponential function equation: P = k4·e^(λA)+d3

[0125] This function is applicable to the case where sandstorm activity explodes exponentially after the exposed area exceeds a critical threshold, where:

[0126] λ is the exponential growth rate coefficient, which represents the rate of sandstorm outbreak;

[0127] k4 is the proportional coefficient;

[0128] d3 is a constant term;

[0129] Recommended parameter ranges: k4∈[0.01,0.1], λ∈[0.005,0.02], d3∈[0.2,0.4].

[0130] The parameter value is related to factors such as regional sand sensitivity, surface material composition, and vegetation coverage.

[0131] Model parameter determination method:

[0132] Data partitioning: historical data is divided into three intervals according to the SWAD threshold;

[0133] Independent fitting of each interval: Use nonlinear least squares method to fit the model parameters of each interval;

[0134] Parameter optimization: Bayesian optimization method is used to minimize the prediction error;

[0135] Model validation: Use cross-validation to evaluate the model's predictive ability in each interval;

[0136] Parameter adjustment: fine-tune parameters based on verification results and test the smoothness of interval transitions;

[0137] (3) Determination of threshold value and conversion method of wind and sand response model:

[0138] ①Threshold determination method:

[0139] A data-driven approach is used to determine the two key thresholds of SWAD_L and SWAD_H:

[0140] a. Data preparation: Collect observational data on wind and sand activity for at least three complete hydrological years, including FD, BD, DS, and TSD. Divide the dataset by season (spring, summer, autumn, and winter), ensuring that the sample size for each season is ≥12 periods.

[0141] b. Multimodal distribution detection:

[0142] Perform kernel density estimation (KDE) on the SWAD data series of each season;

[0143] Analyze the KDE curve morphology and identify multi-peak features;

[0144] The Hilbert-Huang transform (HHT) is used to highlight the nonlinear features in SWAD data;

[0145] c. Segment cluster analysis:

[0146] Gaussian mixture model (GMM) fitting was performed using the expectation maximization (EM) algorithm;

[0147] The Bayesian Information Criterion (BIC) is used to determine the optimal number of clusters, which generally ranges from 3 to 5;

[0148] When the BIC value decreases by <5%, a smaller number of clusters is selected;

[0149] When the BIC difference between different cluster numbers is less than 10, the clustering result with clearer physical meaning is selected;

[0150] Calculate the statistical characteristics of each cluster (mean, standard deviation, quantile);

[0151] d. Jump point detection:

[0152] The jump detection is performed based on the pruning exact linear time algorithm PELT (Pruned Exact Linear Time) and the change point detection algorithm. SWAD_L is defined as the SWAD value of the first significant jump point, SWAD_H is defined as the SWAD value of the second significant jump point, and SWAD_H>SWAD_L

[0153] e. Threshold verification:

[0154] Use historical sandstorm disaster records to verify the rationality of the threshold;

[0155] Calculate the probability change rate of wind and sand activity at each threshold, requiring the change rate to be ≥40%;

[0156] If the verification fails, adjust to the corresponding quantile value based on the probability distribution of wind and sand activity:

[0157] SWAD_L is adjusted to approximately the 75th percentile; SWAD_H is adjusted to approximately the 90th percentile.

[0158] ②Implementation of Jump Detection Algorithm:

[0159] Algorithm principle: Based on the modified PELT (Pruned Exact Linear Time) and enhanced change point detection method, it is used to identify the nonlinear structure of SWAD time series.

[0160] Mathematical expression:

[0161] Cost function: \(C(SWAD_{1:n})=\sum_{i = 1}^{m + 1}[L(SWAD_{\tau_{i - 1}+1:\tau_i})+\eta]\);

[0162] Where \(L\) is the negative log-likelihood function, \(\eta\) is the complexity penalty term, \(\tau\) is the change point position, and \(m\) is the number of change points;

[0163] Optimal segmentation search: \(F(n)=\min_{m,\tau1,...,\tau}\) m}\(C(SWAD_{1:n})\);

[0164] Pruning strategy: If for any \(k < t < s\), \(F(t)+C(SWAD_{t + 1:s})\geq F(s)\), then \(t\) can be pruned from the candidate change points after \(k\);

[0165] Program example, algorithm pseudocode:

[0166] function jump_detection_algorithm(SWAD_series,penalty = 0.05,min_segment_length = 5,significance_level = 0.01):

[0167] """

[0168] Implement the jump detection algorithm for SWAD sequences <​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​

[0180] cost[0]=-penalty#boundary condition

[0181] #Optimal segment record

[0182] cp={}#Save the optimal change point at each position

[0183] cp[0]=[]

[0184] #Pruning set

[0185] R={0}#Search starts from position 0

[0186] #Main loop

[0187] for s in range(1,n+1):

[0188] #Calculate the cost function

[0189] candidates = {}

[0190] for t in R:

[0191] if st <min_segment_length:

[0192] continue

[0193] #Calculate the negative log-likelihood of the current segment

[0194] segment_data=SWAD_series[t:s]

[0195] segment_cost=neg_log_likelihood(segment_data)

[0196] candidates[t]=cost[t]+segment_cost+penalty

[0197] #Find the optimal previous change point

[0198] t_star=min(candidates,key=lambda t:candidates[t])

[0199] cost[s] = candidates[t_star]

[0200] cp[s]=cp[t_star]+[t_star]

[0201] #Prune the search space

[0202] R_new={s} #Always contains the latest position

[0203] for t in R:

[0204] if cost[t]+min_cost_possible(SWAD_series[t:]) <cost[s]:

[0205] R_new.add(t)

[0206] R=R_new

[0207] #Extract jumping points

[0208] jump_points=cp[n]

[0209] #Significance test

[0210] validated_jumps=[]

[0211] for jump in jump_points:

[0212] #Calculate the SWAD mean before and after the jump point

[0213] pre_jump_mean=mean(SWAD_series[max(0,jump-min_segment_length):jump])

[0214] post_jump_mean=mean(SWAD_series[jump:min(n,jump+min_segment_length)])

[0215] #Perform t-test

[0216] t_stat,p_value=ttest_ind(

[0217] SWAD_series[max(0,jump-min_segment_length):jump],

[0218] SWAD_series[jump:min(n,jump+min_segment_length)])

[0219] #Calculate the percentage of change

[0220] change_percent=abs((post_jump_mean-pre_jump_mean) / pre_jump_mean)*100

[0221] #Only keep significant jump points with large changes

[0222] if p_value<significance_level and change_percent> =40:

[0223] validated_jumps.append((jump,change_percent,p_value))

[0224] # Sort by change

[0225] validated_jumps.sort(key=lambda x:x[1],reverse=True)

[0226] #Return the jump point location and its SWAD value

[0227] return[(jump,SWAD_series[jump])for jump,_,_in validated_jumps]

[0228] function neg_log_likelihood(data):

[0229] """

[0230] Compute the negative log-likelihood of a data segment

[0231] parameter:

[0232] data: data sequence

[0233] return:

[0234] nll: negative log-likelihood

[0235] """

[0236] #Assume that the data follows a normal distribution

[0237] mu=mean(data)

[0238] sigma = std(data)

[0239] n=len(data)

[0240] #Calculate negative log-likelihood

[0241] #L=-n / 2*log(2π)-n / 2*log(σ 2 )-1 / (2σ 2 )*Σ(Xi-μ) 2

[0242] nll=n / 2*log(2*π)+n / 2*log(sigma**2)+sum((data-mu)**2) / (2*sigma**2)

[0243] return nll

[0244] function min_cost_possible(data):

[0245] """

[0246] Calculate the minimum possible cost of a data segment

[0247] parameter:

[0248] data: data sequence

[0249] return:

[0250] min_cost: minimum possible cost

[0251] """

[0252] #The minimum possible cost is to assume that the entire sequence is a segment

[0253] return neg_log_likelihood(data)

[0254] ```

[0255] Algorithm parameter settings:

[0256] Complexity penalty parameter (η): winter: 0.035, spring: 0.040, summer: 0.030, autumn: 0.035;

[0257] Minimum segment length: 5 data points (about 50 days) to ensure that the detected change points have time persistence;

[0258] Significance level: p < 0.01, requiring the SWAD values ​​before and after the change point to be significantly different;

[0259] Change amplitude requirement: ≥40%, ensuring that the detected change points have actual physical significance;

[0260] Post-processing method:

[0261] Time series smoothing: LOESS (locally weighted regression scatter point smoothing) is used to reduce noise interference;

[0262] Mutation point clustering: Mutation points with a distance of less than 30 days are merged;

[0263] Seasonal adjustment: Dynamically adjust the detection threshold by ±10% based on seasonal characteristics;

[0264] ③Model conversion smoothing:

[0265] In order to avoid jumps when switching between adjacent interval models, a weighted transition zone design is adopted:

[0266] a. Set a transition interval of ±0.5 SWAD units on both sides of each segment point (SWAD_L, SWAD_H);

[0267] b. The probability of wind and sand is calculated using a weighted fusion method in the transition zone:

[0268] Taking SWAD_L as an example, when SWAD∈[SWAD_L-0.5,SWAD_L+0.5]:

[0269] P = w1·P_mild+w2·P_moderate;

[0270] Where weight w1 = (SWAD_L + 0.5 - SWAD) / 1.0, w2 = 1 - w1;

[0271] c. Smooth transition ensures the continuity of the first-order derivative of the model at the segmentation point, avoiding unreasonable jumps in the prediction results;

[0272] (4) Parameter optimization method of wind and sand response model:

[0273] ① Parameter initialization:

[0274] The initial values ​​of the model parameters in each interval are obtained by piecewise linear regression of historical data, and the initial fitting R 2 ≥0.75;

[0275] ②Parameter optimization algorithm:

[0276] The Bayesian optimization framework is used to determine the optimal parameter set for each model:

[0277] a. Objective function: Minimize the logarithmic loss between the predicted probability of sandstorms and the actual observations:

[0278] L(θ)=-∑[y_i·log(P_i)+(1-y_i)·log(1-P_i)]

[0279] Where θ represents the model parameter set, y_i is the actual observed sandstorm event (0 or 1), and P_i is the model prediction probability

[0280] b. Bayesian optimization process:

[0281] Set parameter prior distribution: Set reasonable prior distribution for each parameter based on physical meaning;

[0282] Gaussian process regression is used to construct a probabilistic proxy model of the objective function;

[0283] Determine the next set of parameters to be evaluated by maximizing the expected improvement (EI) acquisition function;

[0284] Iterate the optimization until convergence (improvement is <1% for 5 consecutive iterations or the maximum number of iterations is reached);

[0285] c. Cross-validation: 10-fold cross-validation is used to evaluate parameter robustness, requiring the performance standard deviation on each fold validation set to be <10%;

[0286] ③Seasonal parameter adjustment:

[0287] The model parameters are adjusted according to the season, and the seasonal adjustment coefficient S_adj is introduced:

[0288] a. Winter (December-February): keep the original parameters unchanged, S_adj = 1.0;

[0289] b. Spring (March-May): The sandstorm activity intensifies, and the nonlinear coefficient increases, k2_spring = k2·1.15, λ_spring = λ·1.20, and other parameters are multiplied by S_adj = 1.05;

[0290] c. Summer (June-August): The period of weakened wind and sand activity, multiply each parameter by S_adj=0.85;

[0291] d. Autumn (September-November): The turning point of wind and sand activity. Each parameter is multiplied by S_adj=0.95.

[0292] ④Model evaluation indicator system:

[0293] a. Precision: The ratio of correctly predicted sandstorm events to all predicted events, which must be ≥ 0.80;

[0294] b. Recall: The ratio of correctly predicted sandstorm events to all actual events, required to be ≥ 0.75;

[0295] c. F1 value: the harmonic mean of precision and recall, required to be ≥ 0.75;

[0296] d.AUC value: area under the ROC curve, required to be ≥0.85;

[0297] e. Blair Skill Score (BSS): The prediction skill relative to the climate mean, required to be ≥ 0.40.

[0298] (5) The gradient boosted decision tree (GBDT) model is used to integrate seasonal factors, meteorological conditions, and the exposed area of ​​sandbars to predict the probability of sandstorm activity. The model prediction accuracy AUC value must be ≥0.85.

[0299] (6) Application example of the three-stage wind-sand response model:

[0300] Taking the Shannan section of the middle reaches of the Yarlung Zangbo River as an example, combined with measured data from 2018 to 2020, the model construction and application process is as follows:

[0301] ①SWAD calculation and threshold determination:

[0302] Based on the wind and sand event data recorded by the local weather station, the SWAD sequence is calculated and determined by the jump detection algorithm:

[0303] SWAD_L=8.5 (mild to moderate cutoff point);

[0304] SWAD_H=13.2 (moderate to severe cutoff point);

[0305] ② Optimization results of model parameters in each interval:

[0306] Mild range (SWAD≤8.5):

[0307] k1=0.0025,d1=0.08,goodness of fit R 2 =0.86;

[0308] Moderate range (8.5 <SWAD≤13.2):

[0309] k2=0.00012, k3=0.0015, d2=0.15, goodness of fit R 2 =0.92;

[0310] Severe range (SWAD>13.2):

[0311] k4 = 0.035, λ = 0.011, d3 = 0.28, goodness of fit R 2 =0.94;

[0312] ③Model prediction verification:

[0313] During the peak sandstorm season in spring 2020 (March to May):

[0314] Actual observed incidence of wind-blown sand events: 56.7%;

[0315] The probability of wind and sand predicted by the three-stage model is 52.3%, with a prediction error of 7.8%;

[0316] Single linear model prediction: 42.1%, prediction error 25.7%;

[0317] Further evaluation indicators:

[0318] Precision: 0.83; Recall: 0.79; F1 value: 0.81; AUC: 0.88; BSS: 0.45;

[0319] The verification results show that the three-stage model can accurately capture the nonlinear outbreak characteristics of high wind and sand periods, and is significantly better than the traditional single model.

[0320] 4. Dynamic calculation and regulation of minimum ecological flow

[0321] (1) Determination of critical exposed area:

[0322] Based on the wind-sand response model, the critical exposed area A_critical that is acceptable for wind-sand risk is determined;

[0323] In the light sandstorm range, the maximum A value with P≤0.3 is taken as A_critical_light;

[0324] In the moderate sandstorm range, the maximum A value with P≤0.5 is taken as A_critical_medium;

[0325] In the severe sandstorm area, the maximum A value with P≤0.7 is taken as A_critical_heavy;

[0326] Select the corresponding critical area threshold according to the SWAD value.

[0327] (2) Calculation of minimum ecological flow:

[0328] Substitute A_critical into the inverse AQ relationship model in step 2 to solve the minimum ecological flow Q_eco that meets the needs of wind and sand suppression. The required flow is dynamically adjusted with the season and wind and sand intensity to form an adaptive minimum ecological flow curve.

[0329] (3) Adaptive dynamic sliding window algorithm:

[0330] ① Window length design principle:

[0331] The sliding window length is dynamically adjusted according to the intensity of wind and sand activity, ranging from 2 to 4 months with a step length of 10 days.

[0332] Window length (WL) calculation formula:

[0333] WL = 2 + 2×(SWAD - SWAD_min) / (SWAD_max - SWAD_min) (months);

[0334] Where:

[0335] SWAD is the current seasonal wind-sand activity characteristic value;

[0336] SWAD_min is the minimum SWAD value observed historically;

[0337] SWAD_max is the maximum SWAD value observed historically;

[0338] When SWAD ≤ SWAD_L, WL takes the lower limit value of 2 months;

[0339] When SWAD ≥ SWAD_H, WL takes the upper limit value of 4 months;

[0340] When SWAD_L < SWAD < SWAD_H, WL is calculated by linear interpolation;

[0341] ② Window sliding and update strategy:

[0342] a. The basic step size is 10 days, and it slides forward one step size each time;

[0343] b. Calculate the average SWAD value within the new window and adjust the window length accordingly;

[0344] c. When the change in the minimum ecological flow calculated twice consecutively exceeds 20%, reduce the step size to 5 days to improve the time resolution;

[0345] d. When the change in the minimum ecological flow calculated 4 times consecutively is less than 5%, increase the step size to 15 days to reduce the calculation amount;

[0346] ③ Change point detection trigger mechanism:

[0347] a. When the change trends of Q_eco are opposite twice consecutively and the amplitudes both exceed 15%, trigger the change point detection;

[0348] b. Determine whether it is a real change in hydrometeorological conditions or an abnormal fluctuation;

[0349] c. When it is determined to be a real change, start the manual verification process;

[0350] d. When it is determined to be an abnormal fluctuation, use the smoothing processing technology to reduce the influence of the fluctuation;

[0351] (4) Manually adjust the runoff of the fixed measuring point section of the river channel to above the minimum ecological flow, and update the model data for optimization in real time. And through real-time monitoring of the changes in wind-sand activities after application, continuously optimize the minimum ecological flow calculation model:

[0352] ① Establish an application effect evaluation system to evaluate the improvement of wind and sand activities after the implementation of ecological flow regulation on a quarterly basis:

[0353] a. Define improvement index: SWAD improvement rate = (SWAD_before - SWAD_after) / SWAD_before × 100%;

[0354] b. Sandbar area control rate = (A_before-A_after) / A_before×100%;

[0355] c. Sandstorm event probability reduction rate = (P_before-P_after) / P_before×100%;

[0356] ②Trigger update conditions:

[0357] a. When the SWAD value is monitored to drop by more than 15% for two consecutive quarters, the model parameter automatic update process is triggered;

[0358] b. When the cumulative amount of new observation data exceeds 25% of the original training set, a comprehensive update is performed;

[0359] c. When environmental conditions change significantly (e.g., completion of large-scale construction projects in the basin), mandatory updating is required;

[0360] ③ Bayesian Update Framework:

[0361] Calibrate the nonlinear response model parameters using the latest monitoring data:

[0362] a. Bayesian formula: P(θ|D_new,D_old)∝P(D_new|θ)·P(θ|D_old)

[0363] in:

[0364] θ represents the model parameter set (a, b, c, ω, φ, etc.);

[0365] D_new is the newly acquired monitoring data;

[0366] D_old is the historical training data;

[0367] P(θ|D_old) is the prior distribution of the parameters;

[0368] P(D_new|θ) is the likelihood function;

[0369] P(θ|D_new,D_old) is the updated parameter posterior distribution;

[0370] b. Implementation method: Use Markov Chain Monte Carlo (MCMC) method to sample the posterior distribution;

[0371] Use the Metropolis-Hastings algorithm to construct a Markov chain;

[0372] Discard the first 20% of the samples as a burn-in period;

[0373] Retain the last 80% of the samples to calculate the posterior mean and 95% confidence interval of the parameter;

[0374] c. Parameter update constraints:

[0375] The change rate of any parameter in a single update does not exceed ±20%;

[0376] The updated parameter combination must satisfy the physical meaning constraints;

[0377] Goodness of fit R of the updated model 2 Not less than 95% of the original model;

[0378] ④ Dynamic adjustment of SWAD threshold:

[0379] a. Dynamically adjust the SWAD_L and SWAD_H thresholds as wind and sand activity improves;

[0380] b. Update the posterior distribution of the threshold using a similar Bayesian framework;

[0381] c. The threshold change rate is limited to ±10% / year to ensure system stability;

[0382] ⑤Update verification mechanism:

[0383] a. Build an independent validation dataset (latest 20% data);

[0384] b. Evaluation indicators include RMSE, MAE, R 2 and forecast bias;

[0385] c. The new model is only officially applied when the prediction accuracy improves by more than 5%;

[0386] d. Otherwise, keep the original model, incorporate the new data into the training set, and wait for the next update.

[0387] (5) Scope of application and restrictions

[0388] Applicable river section types:

[0389] Seasonal wind and sand activities in the mid- and low-latitude (15°-45°) regions affect river sections;

[0390] The riverbed is mainly sandy, with significant seasonal sandbar exposure;

[0391] Wide and shallow rivers with a river width / water depth ratio of >20;

[0392] River sections with annual flow variations >300%;

[0393] Data requirements:

[0394] Remote sensing data: spatial resolution ≤ 30m, temporal resolution ≤ 1 month, continuous observation ≥ 3 years;

[0395] Flow data: time resolution ≤ 10 days, measurement error ≤ ± 3%;

[0396] Meteorological data: including at least three indicators: wind speed, visibility, and PM10 concentration;

[0397] Application Restrictions:

[0398] It is not applicable to high-altitude river sections with ice and snow cover period of more than 2 months;

[0399] It is not applicable to river sections without obvious seasonal sandbar exposure features;

[0400] It is not applicable to areas where there is no significant correlation between wind and sand activities and exposed river channel area;

[0401] River sections that are heavily regulated by water conservancy projects need to add a human factor correction module;

[0402] The model applies boundary conditions:

[0403] The SWAD index system is applicable to areas where the annual change in the probability of wind and sand activity is ≥200%;

[0404] The AQ relationship model is applicable to the sandbar area response curve R 2 River sections with a value of ≥0.75;

[0405] Expert intervention is required when encountering extreme climate events (once-in-a-century rainstorms, severe sandstorms).

[0406] Compiling the above-mentioned method steps into a program and then storing it on a hard disk or other non-transitory storage medium constitutes an embodiment of the present invention's "a non-transitory readable recording medium"; and electrically connecting the storage medium to a computer processor and being able to complete the regulation of the river ecological flow through data processing constitutes an embodiment of the present invention's "a system for regulating the river ecological flow."

[0407] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computers containing computer-usable program code, or on available storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.).

[0408] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0409] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0410] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0411] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for regulating river ecological flow, characterized in that: It includes the following steps: S1. Collect time series data of the exposed area of sandbars in the river channel, the runoff of the fixed measuring point section, and the probability of sand and wind activities; S2. Establish a function model between the exposed area of the sandbar and the runoff of the fixed measuring point section at the same moment, and use the particle swarm algorithm to optimize the parameters of the function model; S3. Establish a sand and wind activity intensity model, determine the intermediate value of the two sand and wind activity intensities through the jump detection algorithm, and divide the sand and wind activity intensity into three intervals: mild, moderate, and severe by the two intermediate values; among them, the jump detection algorithm is synthesized based on the trimmed linear time exact algorithm and the change point detection algorithm, and is used for the non-linear structure identification of the time series of sand and wind activity intensity values; S4. Establish a sand and wind response model for describing the relationship between the probability of sand and wind activities and the exposed area of the sandbar in the same period for the three intervals respectively. For the continuous sampling of time series data involved in the function model, use the adaptive dynamic sliding window algorithm, and the window length is dynamically adjusted according to the sand and wind activity intensity, with a range of 2-4 months and an initial sliding step of 10 days; [[ID=⑤]]S5. Determine the critical exposed area of the sandbar according to the set threshold of the probability of sand and wind activities in the sand and wind response model, substitute the critical exposed area into the function model in step S2, and calculate the corresponding runoff of the fixed measuring point section as the minimum ecological flow; adjust the sliding step according to the change range of the minimum ecological flow within the set time period; [[ID=⑥]]S6. Manually adjust the runoff of the fixed measuring point section of the river channel to above the minimum ecological flow, and update the model data for optimization in real time.

2. A method for regulating river ecological flow according to claim 1, characterized in that: [[ID=⑦]]The sand and wind activity intensity model is: set the sand and wind activity intensity SWAD = α·FD + β·BD + γ·DS + δ·TSD; where, FD is the floating dust frequency, BD is the sand blowing frequency, DS is the sandstorm frequency, TSD is the dust storm duration, and α, β, γ, δ are seasonal weight coefficients, which are determined by the principal component analysis method in the same season; set the intermediate value separating the mild and moderate intervals as SWAD_L, and the intermediate value separating the moderate and severe intervals as SWAD_H; then the three sand and wind response models: [[ID=⑧]]In the mild interval, the sand and wind response model corresponding to the state of SWAD ≤ SWAD_L: [[ID=⑨]]P = k1·A + d1; [[ID=⑩]]In the moderate interval, the sand and wind response model corresponding to the state of SWAD_L < SWAD ≤ SWAD_H: P=k2·A 2 +k3·A+d2; [[ID=⑪]]In the severe interval, the sand and wind response model corresponding to the state of SWAD > SWAD_H: P=k4·e λA +d3: [[ID=⑫]]Among them, λ is the exponential growth rate coefficient, representing the rate of sand and wind outbreaks, k1, k2, k3, k4 are proportionality coefficients, d1, d2, d3 are constant terms, which are optimized by the Bayesian algorithm, P is the probability of sand and wind activities, A is the exposed area of the sandbar, and belongs to the observed data.

3. A method for regulating river ecological flow according to claim 2, characterized in that: [[ID=⑬]]The function model between the exposed area of the sandbar and the runoff of the fixed measuring point section in step S2 is: When the exposed area of ​​the sandbar A≤ζ·A sand When the sandbar is completely submerged, it is determined that there is no need to adjust the ecological flow by default; Set the threshold of seasonal exposed sandbar area When ζ·A sand <A≤A thres When the cross-sectional runoff at the fixed measuring point is Q=Q base +κ·(A-ζ·A sand ); If A>A thres , fixed measuring point cross-sectional runoff [[ID=⑭]]Where: A sand : the total area of ​​exposed sandbars, determined by the maximum historical exposed area extracted from remote sensing images; [[ID=⑮]]ζ: Submergence threshold coefficient, default value is 0.02, optimization range is 0.01-0.05; Q base : Basic ecological flow, taking the 90% quantile of the historical minimum monthly average flow; [[ID=⑯]]κ: Flow compensation coefficient, determined by linear regression; [[ID=⑰]]a is the maximum exposed area coefficient, representing the theoretical maximum exposed area in the zero flow state; [[ID=⑱]]b is the attenuation coefficient, representing the reduction rate of the exposed area when the flow increases; c is the shape parameter, which determines the nonlinearity of the function curve; μ is the average baseline exposed area; ω is the seasonal variation coefficient, which represents the fluctuation range of exposed area relative to the reference area; φ is the phase parameter, which characterizes the time offset of seasonal variation; t is the time point of observation, and the serial number sorted from 1 to 365 within the year is used as the t value; Parameters a, b, c, μ, ω, and φ are optimized by particle swarm optimization, and their hyperparameters are: Goodness of fit R 2 ≥0.90; the root mean square error of the sandbar exposed area A is less than 10% of the average sandbar exposed area.

4. A method for regulating river ecological flow according to claim 3, characterized in that: The steps for collecting time series data on the exposed area of ​​sandbars in a river channel include: The spatial resolution of the selected remote sensing image should be ≤30×30m 2 , covering at least three hydrological years including wet, normal and dry years, and collecting remote sensing images of the river area when the cloud cover rate of the image is ≤5%; ENVI software was used to perform radiometric calibration and atmospheric correction on remote sensing images. The Normalized Difference Vegetation Index (NDVI) was combined with supervised classification using a deep residual convolutional neural network to extract the vector boundaries of exposed sandbars at each period and calculate the exposed sandbar area A. NDVI = (NIR-R) / (NIR+R), where NIR is the reflectance of the near-infrared band and R is the reflectance of the red band. The criterion for identifying exposed sandbars is NDVI < 0.2 and the terrain elevation is between the lowest water level and the flood level of the river. Field measurements were performed to verify whether the accuracy of the exposed sandbar area extraction met the requirements.

5. A method for regulating river ecological flow according to claim 4, characterized in that: The deep residual convolutional neural network is based on the ResNet-50 architecture, which includes 5 residual blocks and a total of 50 convolutional layers. A channel attention module is added to the residual block to improve sensitivity to sandbar features. A feature pyramid structure is adopted, combined with feature maps of different scales, to improve the detection ability of sandbars of different sizes; a boundary refinement network is added to accurately extract sandbar boundaries; a combination of weighted cross entropy loss and boundary-aware loss is used to improve the accuracy of sandbar boundary recognition; among them, the neural network training parameters include the learning rate: the initial value is 0.001, which is adjusted by the cosine annealing strategy; an adaptive moment estimation optimizer is used, the exponential decay rate of the first-order moment β1=0.9, and the exponential decay rate of the second-order moment β2=0.999; the batch size is 16, and the number of training rounds is initially set at 200 rounds. If there is no improvement in the verification accuracy after 10 rounds, the training will be completed in advance.

6. A method for regulating river ecological flow according to claim 5, characterized in that: When the cloud coverage is >5% but <20%, cloud detection and cloud removal algorithms are used, and time series interpolation is performed on the area under the clouds. When data is missing in a local area, a spatiotemporal fusion algorithm is used to fill the gaps by combining data from adjacent time points. When data is completely missing for a specific period, a Markov chain Monte Carlo method based on historical data from the same period is used to generate simulated data.

7. The method for regulating river ecological flow according to claim 3, characterized in that: The window length in months is: WL=2+2×(SWAD-SWAD_min) / (SWAD_max-SWAD_min); where SWAD_min is the minimum SWAD value of historical observations; SWAD_max is the maximum SWAD value of historical observations; when the change in the minimum ecological flow calculated twice in succession exceeds 20%, the step length is reduced to 5 days; when the change in the minimum ecological flow calculated four times in succession is less than 5%, the step length is increased to 15 days; when the change trends of the minimum ecological flow calculated twice in succession are opposite and the amplitude exceeds 15%, the change point detection is triggered to determine whether it is a real change in hydrological and meteorological conditions or an abnormal fluctuation. When it is determined to be a real change, the manual verification process is started; when it is determined to be an abnormal fluctuation, the smoothing processing technology is used to reduce the impact of the fluctuation.

8. The method for regulating river ecological flow according to claim 3, characterized in that: It also includes a dynamic model update mechanism: The improvement of wind and sand activity after the implementation of ecological flow regulation is evaluated every quarter; when the SWAD value of wind and sand activity intensity is monitored to drop by more than 15% for two consecutive quarters, the SWAD_L and SWAD_H values ​​and wind and sand response model parameters are automatically adjusted dynamically using the latest monitoring data; through the validation set test, when the prediction accuracy is improved by more than 5%, the model parameters are updated; otherwise, the original model is retained and the new data is included in the training set, waiting for the next update; when the cumulative amount of new observation data exceeds 25% of the original training set or the environmental conditions change significantly, all models are fully updated.

9. A non-transitory readable recording medium for storing one or more programs comprising a plurality of instructions, characterized in that: When the instruction is executed, the processing circuit will be caused to execute the method for regulating river ecological flow according to any one of claims 1 to 8.

10. A system for regulating river ecological flow, comprising a processing circuit and a memory electrically coupled thereto, characterized in that: The memory configuration stores at least one program, the program including a plurality of instructions, and the processing circuit runs the program to execute a method for regulating river ecological flow according to any one of claims 1-8.

Images